Table 5 Ablation study on CyberDetect-MLP components.

From: CyberDetect MLP a big data enabled optimized deep learning framework for scalable cyberattack detection in IoT environments

Model variant

Removed/modified component

Accuracy (%)

Precision (%)

Recall (%)

F1-score (%)

CyberDetect-MLP (Full Model)

—

98.87

98.74

98.62

98.68

MLP-NoFeatureSelect

Removed Mutual Information-based selection

96.22

95.89

95.34

95.61

MLP-NoBatchNorm

Removed Batch Normalization

96.91

96.50

96.07

96.28

MLP-NoDropout

Removed Dropout (p = 0.5)

97.08

96.83

96.32

96.57

MLP-NoScheduler

Constant learning rate (no scheduler)

96.43

96.01

95.46

95.73

MLP-RawFeatures

No preprocessing or normalization

94.86

94.35

93.82

94.08

Back to article page